A method, system, equipment, and medium for multi-sensor data fusion monitoring and early warning in substations.

By defining fault propagation paths and constructing a state space benchmark library in substations, and generating fault propagation path state vectors by combining multidimensional sensor data, the problems of neglecting electrical topology relationships and relying on preset thresholds in existing technologies are solved, thus realizing effective monitoring of fault propagation paths and identification of unknown risks.

CN123069097APending Publication Date: 2026-11-10GUIZHOU POWER GRID CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611145037.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-30
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing multi-sensor monitoring technologies for substations lack consideration of the electrical topology relationships between equipment and rely heavily on preset thresholds and known state model libraries, resulting in an inability to effectively identify complex faults and unknown risks along the fault propagation path.

Method used

Based on the electrical topology of the substation, a fault propagation path is defined, a state vector set is generated, a path state space benchmark library is constructed, the boundary of the unknown risk space and the cluster center of the known fault are calibrated by kernel density estimation, the state vector of the fault propagation path is generated by combining multi-dimensional sensor data, and the fault is identified by spatial attribution determination.

Benefits of technology

Effectively capture the evolution of fault propagation along electrical paths, eliminate monitoring blind spots, accurately identify chain-like complex faults and unknown abnormal modes, and improve the monitoring and early warning capabilities of substations for complex faults and unknown risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069097A_ABST
    Figure CN123069097A_ABST
Patent Text Reader

Abstract

This invention discloses a method, system, equipment, and medium for multi-sensor data fusion monitoring and early warning in substations. The method includes: defining fault propagation paths based on the substation's electrical topology; generating state vector sets based on historical sensor data; calibrating unknown risk space boundaries and known fault cluster centers based on the state vector sets; constructing a path state space benchmark library; collecting real-time multi-dimensional sensor data for monitoring nodes along the fault propagation path; generating fault propagation path state vectors based on the real-time multi-dimensional sensor data; performing state space attribution determination to obtain path state attribution results; and generating a substation fusion monitoring and early warning status. This invention can accurately identify chain-like complex faults evolving along the propagation path, while maintaining effective perception of unknown abnormal patterns outside the data distribution, significantly improving the monitoring and early warning capabilities for complex faults and unknown risks in substations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of monitoring and early warning technology, and in particular to a method, system, equipment and medium for monitoring and early warning of substations using multi-sensor data fusion. Background Technology

[0002] The monitoring field encompasses real-time or periodic detection, data acquisition, analysis, and response to the status of specific systems or processes. Its core content involves utilizing sensor technology to acquire physical quantities, environmental parameters, or equipment operating data, aggregating this information through data transmission networks, and using data analysis to determine whether the system status is normal. This allows for triggering alarms or taking control actions when anomalies occur. This technology is widely used in industrial automation, power system operation, environmental monitoring, and security. Specifically, multi-sensor data fusion monitoring and early warning in substations involves deploying multiple types of sensors in the substation environment to simultaneously collect operational data from different dimensions. Data alignment and correlation techniques are used to process time-series information from different sensors, merging multi-source data into a unified status representation. Based on pre-set thresholds or a state model library, the fused data is continuously compared. When data deviates from the normal range or conforms to a specific fault mode, an early warning mechanism is activated.

[0003] Current monitoring technologies rely on aligning and fusing multi-source data and making judgments based on preset thresholds or known state model libraries. This mechanism lacks consideration of the electrical topology relationships between devices when processing data, causing data fusion to remain at the representational level. It is difficult to capture the chain propagation process of faults along the physical path. For example, minor anomalies in upstream devices and critical data in downstream devices may be ignored because they do not exceed limits individually, but their combination can already pose a serious threat. More importantly, current technologies heavily rely on prior knowledge, namely thresholds and model libraries. They have almost no ability to identify new fault combinations or unknown risk patterns outside the data distribution, creating monitoring blind spots and failing to effectively warn of sudden or unrecorded abnormal states. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, device, and medium for multi-sensor data fusion monitoring and early warning in substations, which solves the problem that existing multi-sensor monitoring technologies in substations lack consideration of the electrical topology relationships between devices and rely heavily on preset thresholds and known state model libraries, resulting in the inability to effectively identify complex faults that evolve along the fault propagation path.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for multi-sensor data fusion monitoring and early warning in substations, comprising: Based on the electrical topology of the substation, a fault propagation path is defined, a state vector set is generated according to historical sensor data, and the unknown risk space boundary and known fault cluster center are calibrated based on the state vector set to construct a path state space benchmark library. For the monitoring nodes along the fault propagation path, real-time multidimensional sensor data is collected, and a fault propagation path state vector is generated based on the real-time multidimensional sensor data. The path state space reference library is invoked, and a state space attribution determination is performed based on the fault propagation path state vector to obtain the path state attribution determination result. The substation integrated monitoring and early warning status is generated based on the path status attribution determination result.

[0007] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations described in this invention, the step of defining fault propagation paths based on the substation's electrical topology and generating a state vector set based on historical sensor data includes: Extract the electrical connection relationships of each equipment node in the electrical topology of the substation, and define the paths between equipment nodes with direct electrical connections as fault propagation paths; Call the sensor data in the historical database and backtrack to calculate the corresponding historical status readings for all defined fault propagation paths; The historical state readings are integrated into a vector form according to the path identifier to generate a state vector set.

[0008] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations described in this invention, the method for constructing a path state space reference library includes: Based on the state vector set, the probability density distribution of each data point is calculated in the multidimensional state space, and a density threshold is defined according to the set statistical quantile. The spatial region with a density estimate lower than the density threshold is fitted as a continuous geometric boundary to obtain the unknown risk space boundary. For the known fault vectors that have been marked in the state vector set, they are grouped according to the fault category, and the arithmetic mean of the vector coordinates in each group is calculated to obtain the set of geometric center point coordinates for each known fault category. By integrating the unknown risk space boundary and the geometric center point coordinate set, a path state space benchmark library is established.

[0009] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations described in this invention, the method for collecting real-time multi-dimensional sensor data includes: For the monitoring nodes along the fault propagation path, multiple types of sensors are invoked to synchronously collect the operating signals of each node, and multi-source heterogeneous readings containing timestamps are obtained. The multi-source heterogeneous readings are synchronized and aligned according to the acquisition timestamp. All synchronized and aligned signal readings are combined into a unified data structure according to the path node identifier to obtain real-time multidimensional sensing data.

[0010] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations described in this invention, the step of generating a fault propagation path state vector based on the real-time multi-dimensional sensor data includes: The partial discharge signal time series and dissolved gas concentration data in oil of a specified node are extracted from the real-time multidimensional sensing data. The partial discharge change rate is calculated based on the partial discharge signal time series, and the current acetylene concentration reading is extracted. The partial discharge change rate and the current acetylene concentration reading are combined to generate node feature value pairs. The vibration signal time series is extracted from the real-time multidimensional sensing data, the vibration signal is frequency domain transformed, a target frequency band is set, and the vibration energy within the frequency band is calculated. Calculate the feature product value based on the node feature value pair, and sum the feature product value with the vibration energy to obtain the combined feature scalar; The combined feature scalar and the original values ​​in the node feature value pairs are concatenated in a preset order to generate a fault propagation path state vector.

[0011] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations described in this invention, the step of performing state space attribution determination based on the fault propagation path state vector includes: Extract the unknown risk space boundary from the path state space benchmark library, substitute the coordinates of the fault propagation path state vector into the unknown risk space boundary to determine the spatial location, and generate a regional location identifier. In response to the location identifier of the area falling into an unknown risk area, the path status attribution determination result is marked as an unknown fault; In response to the fact that the area location identifier does not fall into the unknown risk area, the geometric center point coordinate set of the known fault categories in the path state space benchmark library is called to calculate the distance between the fault propagation path state vector and each cluster center point, and the known fault category corresponding to the minimum distance is selected as the judgment result.

[0012] As a preferred embodiment of the multi-sensor data fusion monitoring and early warning method for substations according to the present invention, the generation of the substation fusion monitoring and early warning status includes: In response to the path status attribution determination result being an unknown fault marker, an early warning status containing an unknown fault marker is generated; In response to the path status attribution determination result being a known fault category, a preset fault level mapping table is invoked, the corresponding risk level value is found according to the known fault category index, and the known fault category and the risk level are combined into a unified status description to generate a warning status containing the fault category and risk level.

[0013] Secondly, the present invention provides a substation multi-sensor data fusion monitoring and early warning system, comprising: The state library construction module is used to define fault propagation paths based on the electrical topology of the substation, generate state vector sets based on historical sensor data, calibrate unknown risk space boundaries and known fault cluster centers based on the state vector sets, and construct a path state space benchmark library. The state vector generation module is used to collect real-time multidimensional sensor data for monitoring nodes on the fault propagation path, and generate a fault propagation path state vector based on the real-time multidimensional sensor data. The state space attribution determination module is used to call the path state space benchmark library, perform state space attribution determination based on the fault propagation path state vector, and obtain the path state attribution determination result. The monitoring and early warning module is used to generate the substation integrated monitoring and early warning status based on the path status attribution determination result.

[0014] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of a substation multi-sensor data fusion monitoring and early warning method.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of a substation multi-sensor data fusion monitoring and early warning method.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention extracts the electrical topology of the substation and defines the fault propagation path according to the electrical connection sequence of the equipment. By considering multi-sensor monitoring data within the physical connection sequence, it overcomes the shortcomings of traditional data fusion, which only stays at the representation level and ignores the chain-like relationship between equipment. This enables the monitoring process to effectively capture the evolution law of fault propagation along the electrical path. Based on historical state vector sets, this invention uses kernel density estimation to calibrate the unknown risk space boundary outside the historical data distribution. It also combines known fault cluster centers to construct a multi-dimensional state space benchmark library, replacing the traditional method of simply relying on preset thresholds and known model libraries. This provides a quantifiable spatial definition basis for new anomalies and unknown risks, eliminating the monitoring blind spots caused by incomplete coverage of prior knowledge. During real-time monitoring, this invention generates a path state vector by deeply coupling multi-dimensional features such as partial discharge change rate, acetylene concentration, and vibration energy. It also replaces traditional threshold comparison with spatial attribution determination and distance calculation, which can accurately identify chain-like complex faults evolving along the propagation path. At the same time, it maintains effective perception of unknown abnormal patterns outside the data distribution, significantly improving the monitoring and early warning capabilities of complex faults and unknown risks in substations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall process logic of a substation multi-sensor data fusion monitoring and early warning method provided in one embodiment of the present invention.

[0019] Figure 2 The flowchart illustrates the state space attribution determination process of a substation multi-sensor data fusion monitoring and early warning method according to an embodiment of the present invention.

[0020] Figure 3 The flowchart illustrates the monitoring and early warning process of a multi-sensor data fusion monitoring and early warning method for substations, as provided in one embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for multi-sensor data fusion monitoring and early warning in substations is provided, such as... Figure 1 The specific steps shown are as follows: S100: Define fault propagation paths based on the electrical topology of substations, generate state vector sets based on historical sensor data, calibrate unknown risk space boundaries and known fault cluster centers based on state vector sets, and construct a path state space benchmark library. S200: For monitoring nodes along the fault propagation path, collect real-time multi-dimensional sensor data and generate a fault propagation path state vector based on the real-time multi-dimensional sensor data; S300: Call the path state space baseline library, perform state space attribution determination based on the fault propagation path state vector, and obtain the path state attribution determination result; S400: Generate substation integrated monitoring and early warning status based on the path status attribution determination result.

[0023] It should be noted that this invention, by extracting the electrical topology of the substation and defining the fault propagation path according to the electrical connection sequence of the equipment, considers multi-sensor monitoring data within the context of the physical connection sequence. This overcomes the shortcomings of traditional data fusion, which only focuses on the representational level and ignores the chain-like relationships between equipment. This allows the monitoring process to effectively capture the evolutionary patterns of fault propagation along the electrical path. Based on historical state vector sets, this invention uses kernel density estimation to calibrate the spatial boundary of unknown risks outside the historical data distribution. It also constructs a multi-dimensional state space benchmark library by combining known fault cluster centers, replacing the traditional method of relying solely on preset thresholds and known model libraries. This provides a quantifiable spatial definition basis for new anomalies and unknown risks, eliminating monitoring blind spots caused by incomplete prior knowledge coverage. During real-time monitoring, this invention generates path state vectors by deeply coupling multi-dimensional features such as partial discharge change rate, acetylene concentration, and vibration energy. It uses spatial attribution determination and distance calculation to replace traditional threshold comparison, accurately identifying chain-like complex faults evolving along the propagation path. Simultaneously, it maintains effective perception of unknown anomaly patterns outside the data distribution, significantly improving the monitoring and early warning capabilities of complex faults and unknown risks in substations.

[0024] Example 2, refer to Figure 2 and Figure 3 Based on the previous embodiment, this embodiment provides a specific implementation method for a multi-sensor data fusion monitoring and early warning method for substations.

[0025] In this embodiment of the invention, step S100 defines a fault propagation path based on the electrical topology of the substation, generates a state vector set based on historical sensor data, calibrates the unknown risk space boundary and known fault cluster centers based on the state vector set, and constructs a path state space benchmark library; detailed, it includes the following sub-steps A1 to A5: In A1: Extract the electrical connection relationships of each equipment node in the electrical topology of the substation, and define the paths between equipment nodes with direct electrical connections as fault propagation paths; Specifically, the electrical topology of a substation describes the electrical relationships between devices using nodes and connections. For example, a circuit breaker and an instrument transformer are connected in series on the incoming side of the main transformer. The electrical connections between the nodes of each device in the substation's electrical topology are extracted, and the paths between device nodes with direct electrical connections are defined as fault propagation paths, such as the electrical path between the circuit breaker node and the main transformer node.

[0026] In A2: Sensor data from the historical database is retrieved, and the corresponding historical status readings are calculated back for all defined fault propagation paths; the historical status readings are integrated into a vector form according to the path identifier to generate a status vector set. Specifically, the system retrieves sensor data from the historical database, including partial discharge signals, opening and closing coil currents, dissolved gas concentration readings in the oil, and vibration signal readings collected by each node at historical moments. For all defined fault propagation paths, the system backtracks and calculates the corresponding multidimensional state readings, which include the partial discharge change rate, acetylene concentration value, and combined characteristic scalars.

[0027] Specifically, the calculation of the combined feature scalar includes: performing a frequency domain transformation on the vibration signal to calculate the vibration band energy value, and calculating the combined feature scalar based on the partial discharge change rate, acetylene concentration value, and vibration band energy value; wherein, the vibration band energy value is only used as an intermediate calculation quantity to generate the combined feature scalar and is not used as an independent dimension of the multidimensional state reading.

[0028] Specifically, the multidimensional state readings of each device node included in each defined fault propagation path at the same historical moment are arranged according to the electrical connection order of the devices on the path, forming an ordered sequence of state readings for that path at that moment. Each dimension of the reading corresponds to a sensing feature of a node on the path. The ordered sequence of state readings is used as a multidimensional state vector of the path at the corresponding historical moment, and path identification information and timestamp information are added to the vector. The multidimensional state vectors generated by all paths at all historical moments are classified and collected according to the path identification, forming a set of historical state vectors for each path. The set of state vectors of all paths together constitutes a state vector set.

[0029] In A3: Based on the state vector set, the probability density distribution of each data point is calculated in the multidimensional state space, and the density threshold is defined according to the set statistical quantile. The spatial region with a density estimate lower than the density threshold is fitted as a continuous geometric boundary to obtain the unknown risk space boundary. Specifically, the state vector set contains the state vectors of each path at multiple historical moments. The kernel density estimation method is used to calculate the probability density distribution of all data points in the multidimensional state space. The Gaussian kernel function is selected as the kernel function, and the optimal bandwidth parameter is calculated for each dimension according to the Scott criterion. The probability density estimate of each data point under Gaussian kernel weighting is calculated to obtain the density estimate sequence of all data points.

[0030] Specifically, a statistical quantile is set as the density threshold, for example, the 5th percentile is selected as the boundary for low-density regions. Spatial regions with density estimates below the 5th percentile in the density estimate sequence of all data points are marked as low-density regions and fitted with a continuous geometric equation, such as using an ellipse equation to describe the boundary of the low-density region in two-dimensional space, thus obtaining the spatial boundary of the unknown risk area.

[0031] In A4: The known fault vectors that have been labeled in the state vector set are grouped according to the fault category, and the arithmetic mean of the vector coordinates in each group is calculated to obtain the set of geometric center point coordinates for each known fault category. Specifically, for known fault vectors that have been labeled in the historical state vector set, such as a vector group labeled "internal arc fault", data grouping is performed according to their fault category label. The arithmetic mean is calculated for all vector coordinates in each group to obtain the set of geometric center point coordinates for that category.

[0032] In A5: Integrate the coordinate set of unknown risk space boundaries and geometric center points to establish a path state space benchmark library; Specifically, the fitted unknown risk space boundary is stored in the path state space benchmark library as a geometric equation of the risk area. This geometric equation is used to determine whether the state vector coordinates fall into the unknown risk area. The calculated geometric center point coordinates of various known faults are stored in the path state space benchmark library as a fault center point coordinate set. This coordinate set contains the correspondence between each known fault category and its center point coordinates, and is used to calculate the spatial distance between the vector to be determined and each known fault mode. The generated state vector set is stored in the path state space benchmark library as a historical vector index. This index contains the path identifier, timestamp, and fault category label corresponding to each historical vector, and is used for subsequent querying and tracing.

[0033] Specifically, the aforementioned risk area geometric equations, fault center point coordinate sets, and historical vector indexes are integrated into the same database structure to form a path state space benchmark library that includes spatial boundary definitions, known fault reference points, and historical data indexes.

[0034] It should be noted that step S100 above extracts the electrical topology of the substation and defines the fault propagation path according to the electrical connection sequence of the equipment. It establishes the fusion benchmark of multi-sensor monitoring data on the physical connection relationship between the equipment, changing the traditional approach of collecting data from each node in isolation and ignoring the chain relationship. At the same time, it uses kernel density estimation to calibrate the boundary of the unknown risk space outside the distribution of historical data, and constructs a multi-dimensional state space benchmark library in combination with the cluster centers of known faults. It defines the distribution range of known faults and unknown risks in a spatial geometric way, replacing the judgment benchmark of simply relying on preset thresholds and a limited known model library in the traditional technology, fundamentally eliminating the monitoring blind spots caused by incomplete coverage of prior knowledge.

[0035] In this embodiment of the invention, step S200 involves collecting real-time multidimensional sensor data for monitoring nodes along the fault propagation path, and generating a fault propagation path state vector based on the real-time multidimensional sensor data; this includes the following sub-steps B1 to B6: In B1: For monitoring nodes along the fault propagation path, multiple types of sensors are invoked to synchronously collect the operating signals of each node and obtain multi-source heterogeneous readings containing timestamps; Specifically, for the pre-set instrument transformer nodes, circuit breaker nodes, and main transformer nodes along the fault propagation path, ultra-high frequency sensors are called to collect partial discharge signals, Hall sensors are called to collect opening and closing coil currents, and online chromatographic analysis devices are called to collect dissolved gas concentrations in oil. Each sensor starts collecting data synchronously according to a unified sampling cycle, and records the signal readings collected by each sensor along with the corresponding collection timestamps, forming a multi-source heterogeneous reading that includes equipment node identifiers, signal type identifiers, sampling values, and timestamps.

[0036] For example, at 14:05:30, a pulse amplitude of 15mV is recorded. For the circuit breaker node, a Hall sensor is called to collect the current of the opening and closing coils. For example, at 14:05:30, the current value is measured to be 2.1A. For the main transformer node, an online chromatographic analysis device is called to collect the concentration of dissolved gases in the oil, such as an acetylene concentration of 3.2μL / L. The above readings are combined with the corresponding timestamps to obtain multi-source heterogeneous readings containing timestamps.

[0037] In B2: Synchronization and alignment operations are performed on the multi-source heterogeneous readings based on the acquisition timestamps. All synchronized and aligned signal readings are combined into a unified data structure according to the path node identifiers to obtain real-time multidimensional sensing data. Specifically, based on the preset time window length, the multi-source heterogeneous readings collected by different sensors within the same time window are matched and aligned according to their respective timestamps, so that all types of signals of each node within the same time window correspond to the same acquisition time reference; for example, with a time window of 100ms, all readings during the period from 14:05:30 to 14:05:31 are aligned.

[0038] Specifically, all synchronized signal readings are arranged according to the electrical connection order of each device node on the fault propagation path, and the various signal readings corresponding to each node are classified under the node identifier, forming a unified data structure with path nodes as the organization unit. This data structure includes node timestamps, synchronously acquired signal sequences of various types, and device identifiers, which serve as the basic data input for subsequent feature extraction and state vector generation.

[0039] In B3: Extract the partial discharge signal time series and dissolved gas concentration data in oil of the specified node from the real-time multidimensional sensing data, calculate the partial discharge change rate based on the partial discharge signal time series, extract the current acetylene concentration reading, and combine the partial discharge change rate with the current acetylene concentration reading to generate node feature value pairs. Specifically, the partial discharge signal time series readings of the transformer nodes are retrieved from the real-time multidimensional sensing data. Two amplitude data points at adjacent time points are selected and subtracted. The difference is divided by the time interval to obtain the partial discharge change rate. At the same time, the dissolved gas data in the oil of the main transformer nodes are retrieved from the real-time multidimensional sensing data to extract the acetylene concentration reading at the current moment. The calculated partial discharge change rate and the extracted current acetylene concentration reading are combined into a pair of associated feature values, which serve as the node feature value pair of the fault propagation path at the current moment.

[0040] For example, the time series reading of the partial discharge signal of the transformer node is called. For instance, the pulse amplitude of 15mV at 14:05:30 and 12mV at 14:05:29 are extracted and the change rate of 3mV / s is obtained by subtracting them. The current acetylene concentration of 3.2μL / L is extracted from the dissolved gas data in the oil collected by the online chromatographic analysis device of the main transformer node. The change rate and the concentration value are combined to generate node feature value pairs.

[0041] In B4: Extract the vibration signal time series from real-time multidimensional sensing data, perform frequency domain transformation on the vibration signal, set the target frequency band, and calculate the vibration energy within the frequency band; Specifically, the time series data of the box vibration signal is retrieved from the real-time multidimensional sensing data, and a Fourier transform is performed on the time series to convert it from the time domain to the frequency domain, so as to obtain the spectral amplitude distribution of the vibration signal at each frequency component; the upper and lower cutoff frequencies of the target frequency band are set according to the fault diagnosis requirements, the spectral amplitude data within the frequency band is extracted, and the spectral amplitude within the frequency band is integrated, and the integration result is used as the vibration energy value within the frequency band.

[0042] In B5: The feature product value is calculated based on the nodal feature value pairs, and the feature product value is summed with the vibration energy to obtain the combined feature scalar; Specifically, the partial discharge change rate, the current acetylene concentration reading, and the vibration energy value from the node feature value pair are called. Minimum-maximum normalization is performed on all three, mapping each feature value to the [0,1] interval to obtain the normalized partial discharge change rate, normalized acetylene concentration, and normalized vibration energy value. The normalized partial discharge change rate and normalized acetylene concentration are multiplied to obtain the feature product value. The feature product value and the normalized vibration energy value are added together, and the result of the addition operation is used as a combined feature scalar. This combined feature scalar comprehensively reflects the coupling effect of the partial discharge change rate and acetylene concentration, as well as the magnitude of the vibration energy, and is used subsequently to form a component of the fault propagation path state vector together with the original feature values.

[0043] For example, the target frequency band is set to 100Hz to 500Hz. An integral operation is performed on the spectral amplitude within this band; for instance, integrating the square of the spectral amplitude in the 100-500Hz band yields a vibration energy value of 0.85J. Simultaneously, the partial discharge rate of change (3mV / s) and the acetylene concentration (3.2μL / L) from the node eigenvalue pair are retrieved. Minimum-maximum normalization is performed on all three values ​​to obtain the normalized partial discharge rate of change, normalized acetylene concentration, and normalized vibration energy value. The normalized partial discharge rate of change and the normalized acetylene concentration are multiplied to obtain the eigenproduct value; the eigenproduct value is then summed with the normalized vibration energy value to obtain the combined eigenscalar value of 10.45.

[0044] In B6: The combined feature scalar and the original values ​​in the node feature value pairs are concatenated in a preset order to generate the fault propagation path state vector; Specifically, according to the pre-set dimensional arrangement order, the partial discharge change rate in the node feature value pair is taken as the first dimension, the current acetylene concentration reading is taken as the second dimension, and the combined feature scalar is taken as the third dimension. The values ​​of the above three dimensions are sequentially concatenated and spliced ​​to form a fault propagation path state vector containing the original feature value and the coupled feature value.

[0045] It should be noted that step S200 above synchronously collects various types of real-time sensor data for monitoring nodes along the fault propagation path, and deeply integrates multi-dimensional features such as partial discharge change rate, acetylene concentration value and vibration energy, and then splices them in series to generate a path state vector. This vector not only carries the independent operating information of each node, but also contains the chain-like correlation characteristics between upstream and downstream along the fault propagation path, thus improving the isolated data fusion in traditional technology into a path state representation that reflects the fault chain reaction process.

[0046] In this embodiment of the invention, step S300 calls the path state space reference library, performs state space attribution determination based on the fault propagation path state vector, and obtains the path state attribution determination result; such as Figure 2 The detailed steps shown include the following sub-steps C1 to C3: In C1: Extract the unknown risk space boundary from the path state space benchmark library, substitute the coordinates of the fault propagation path state vector into the unknown risk space boundary to determine the spatial location, and generate the area location identifier. Specifically, the unknown risk space boundary stored in the path state space benchmark library is called. This boundary is defined and describes the distribution range of the unknown risk area in the multi-dimensional state space in the form of a geometric equation. The coordinate values ​​of the current fault propagation path state vector are substituted into the geometric equation for calculation. Based on the calculation result, it is determined whether the coordinates are inside the boundary, thereby determining whether the current state vector falls into the unknown risk area, and generating a region location identifier that records the determination result.

[0047] For example, the equation of an ellipse in two-dimensional space is (x-5). 2 / 9+(y-2) 2 =1, and substitute the coordinates of the fault propagation path state vector, such as vector coordinates (7,3), into the geometric equation of the boundary to calculate the expression on the left (7-5). 2 / 9+(3-2) 2 / 4=0.694, which is compared with the boundary threshold 1. If the calculated value is less than 1, it is determined that the area falls inside the boundary and a region location identifier is generated.

[0048] In C2: In response to the regional location identifier falling into an unknown risk area, the path status attribution determination result is marked as an unknown fault; Specifically, the system judges the location of the region. When the location of the region indicates that the coordinates of the current fault propagation path state vector fall inside the unknown risk space boundary, it means that the current state belongs to an unknown pattern outside the historical data distribution and cannot be matched with any known fault category. Therefore, the path state attribution result is directly marked as an unknown fault state, and the current state category matching process ends. The comparison operation with the known fault cluster center is no longer performed.

[0049] In C3: In response to the fact that the area location identifier does not fall into the unknown risk area, the geometric center point coordinate set of the known fault categories in the path state space benchmark library is called, the distance between the fault propagation path state vector and each cluster center point is calculated, and the known fault category corresponding to the minimum distance is selected as the judgment result. Specifically, when the location identifier indicates that the coordinates of the current fault propagation path state vector do not fall into an unknown risk area, the coordinate set of known fault cluster center points stored in the path state space benchmark library is called. The distance value between the current state vector coordinates and the geometric center point coordinates of each known fault category in the library is calculated to obtain a set of distance values. The set of distance values ​​is compared and filtered, and the known fault category corresponding to the smallest distance value is selected as the fault type that best matches the current state. The path state attribution determination result containing the fault category information is generated.

[0050] For example, consider center points A (4,1) and B (6,4), and calculate the Euclidean distance √[(7-4)] between the state vector of the fault propagation path and center point A, given the coordinates (7,3). 2 +(3-1) 2 =√13≈3.606, calculate its Euclidean distance from the center point B √[(7-6)] 2 +(3-4) 2 =√2≈1.414. Combine all the calculated distance values ​​to obtain the Euclidean distance group. If the marker does not fall into the unknown area, call the Euclidean distance group, filter the smallest distance value of 1.414, and take the known fault category B corresponding to this distance value as the judgment result to generate the path status attribution judgment.

[0051] It should be noted that step S300 above substitutes the coordinates of the fault propagation path state vector into the path state space reference library, and uses spatial location attribution determination and distance calculation with each known fault cluster center to replace the traditional threshold comparison method. This allows the determination process to consider the spatial positional relationship between the current vector and all known fault modes and unknown risk areas in the global state space. It can accurately identify abnormal states that match known fault modes, and also effectively perceive new abnormal modes that fall into the unknown risk boundary.

[0052] In this embodiment of the invention, step S400 generates a substation integrated monitoring and early warning status based on the path status attribution determination result; such as Figure 3 The detailed steps shown include the following sub-steps D1 to D2: In D1: In response to the path status attribution determination result being an unknown fault marker, a warning status containing an unknown fault marker is generated; Specifically, the path status attribution determination result is obtained, and the status identifier field in the determination result is checked to see if it is an unknown fault mark. When the determination result is confirmed to be an unknown fault mark, it indicates that the currently monitored status belongs to a new abnormal mode in an unknown risk area. The unknown fault mark is directly used as the content of the early warning status, and a substation integrated monitoring early warning status containing the unknown fault mark is generated so that operation and maintenance personnel can know that there is an abnormal status that cannot be classified.

[0053] In D2: In response to the path status attribution determination result being a known fault category, the preset fault level mapping table is called, the corresponding risk level value is found according to the known fault category index, and the known fault category and risk level are combined into a unified status description to generate a warning status containing fault category and risk level. Specifically, the path status attribution determination result is obtained. When the determination result contains known fault category information, a preset fault level mapping table is called. The mapping table stores the correspondence between each known fault category and risk level value. The known fault category in the determination result is used as an index to look up the corresponding risk level value in the mapping table. Then, the known fault category and its corresponding risk level value are combined into unified status description information to generate a substation integrated monitoring and early warning status containing fault category and risk level value.

[0054] For example, if the identifier value is "UNKNOWN", an unknown fault status identifier is generated if the condition is met; otherwise, a preset fault level mapping table is called. The fault level mapping table stores the correspondence between fault categories and risk levels. Based on the known fault category index in the judgment, such as category B, the corresponding risk level value of 2 is found to obtain the known fault risk level. Further judgment includes status identifier and known fault category information. If it contains an unknown fault status identifier, the identifier is directly marked as a warning status; otherwise, the known fault risk level 2 is combined with the corresponding fault category B to form a unified status description "Class B Fault Risk Level 2", generating a substation integrated monitoring warning status.

[0055] It should be noted that step S400 distinguishes between known and unknown faults based on the judgment result. For known faults, it obtains the quantified risk level from the fault level mapping table. For unknown faults, it directly marks the unknown fault status. Finally, it generates a fusion monitoring and early warning status that includes fault qualitative information and risk level, providing clear and quantified early warning basis for operation and maintenance personnel and comprehensively improving the accuracy and comprehensiveness of substation monitoring and early warning.

[0056] Example 3: This example provides a substation multi-sensor data fusion monitoring and early warning system, including: The state library construction module is used to define fault propagation paths based on the electrical topology of substations, generate state vector sets based on historical sensor data, calibrate unknown risk space boundaries and known fault cluster centers based on the state vector sets, and construct a path state space benchmark library. The state vector generation module is used to collect real-time multidimensional sensor data for monitoring nodes along the fault propagation path and generate fault propagation path state vectors based on the real-time multidimensional sensor data. The state space attribution determination module is used to call the path state space benchmark library, perform state space attribution determination based on the fault propagation path state vector, and obtain the path state attribution determination result. The monitoring and early warning module is used to generate the substation integrated monitoring and early warning status based on the path status attribution determination results.

[0057] It should be noted that the technical solution of the substation multi-sensor data fusion monitoring and early warning system is based on the same concept as the technical solution of the aforementioned substation multi-sensor data fusion monitoring and early warning method. For details not described in detail in the technical solution of the substation multi-sensor data fusion monitoring and early warning system in this embodiment, please refer to the description of the technical solution of the aforementioned substation multi-sensor data fusion monitoring and early warning method.

[0058] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0059] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through WiFi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a multi-sensor data fusion monitoring and early warning method for substations. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0060] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method proposed in the above embodiments.

[0061] The storage medium proposed in this embodiment belongs to the same inventive concept as the method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0062] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute the method of the embodiments of the present invention.

[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the present invention.

Claims

1. A method for multi-sensor data fusion monitoring and early warning in substations, characterized in that, include: Based on the electrical topology of the substation, a fault propagation path is defined, a state vector set is generated according to historical sensor data, and the unknown risk space boundary and known fault cluster center are calibrated based on the state vector set to construct a path state space benchmark library. For the monitoring nodes along the fault propagation path, real-time multidimensional sensor data is collected, and a fault propagation path state vector is generated based on the real-time multidimensional sensor data. The path state space reference library is invoked, and a state space attribution determination is performed based on the fault propagation path state vector to obtain the path state attribution determination result. The substation integrated monitoring and early warning status is generated based on the path status attribution determination result.

2. The substation multi-sensor data fusion monitoring and early warning method as described in claim 1, characterized in that, The method of defining fault propagation paths based on the substation electrical topology and generating state vector sets based on historical sensor data includes: Extract the electrical connection relationships of each equipment node in the electrical topology of the substation, and define the paths between equipment nodes with direct electrical connections as fault propagation paths; Call the sensor data in the historical database and backtrack to calculate the corresponding historical status readings for all defined fault propagation paths; The historical state readings are integrated into a vector form according to the path identifier to generate a state vector set.

3. The substation multi-sensor data fusion monitoring and early warning method as described in claim 2, characterized in that, The constructed path state space benchmark library includes: Based on the state vector set, the probability density distribution of each data point is calculated in the multidimensional state space, and a density threshold is defined according to the set statistical quantile. The spatial region with a density estimate lower than the density threshold is fitted as a continuous geometric boundary to obtain the unknown risk space boundary. For the known fault vectors that have been marked in the state vector set, they are grouped according to the fault category, and the arithmetic mean of the vector coordinates in each group is calculated to obtain the set of geometric center point coordinates for each known fault category. By integrating the unknown risk space boundary and the geometric center point coordinate set, a path state space benchmark library is established.

4. The substation multi-sensor data fusion monitoring and early warning method as described in claim 3, characterized in that, The real-time multidimensional sensing data collected includes: For the monitoring nodes along the fault propagation path, multiple types of sensors are invoked to synchronously collect the operating signals of each node, and multi-source heterogeneous readings containing timestamps are obtained. The multi-source heterogeneous readings are synchronized and aligned according to the acquisition timestamp. All synchronized and aligned signal readings are combined into a unified data structure according to the path node identifier to obtain real-time multidimensional sensing data.

5. The substation multi-sensor data fusion monitoring and early warning method as described in claim 4, characterized in that, The generation of the fault propagation path state vector based on the real-time multidimensional sensor data includes: The partial discharge signal time series and dissolved gas concentration data in oil of a specified node are extracted from the real-time multidimensional sensing data. The partial discharge change rate is calculated based on the partial discharge signal time series, and the current acetylene concentration reading is extracted. The partial discharge change rate and the current acetylene concentration reading are combined to generate node feature value pairs. The vibration signal time series is extracted from the real-time multidimensional sensing data, the vibration signal is subjected to frequency domain transformation, a target frequency band is set, and the vibration energy within the frequency band is calculated. Calculate the feature product value based on the node feature value pair, and sum the feature product value with the vibration energy to obtain the combined feature scalar; The combined feature scalar and the original values ​​in the node feature value pairs are concatenated in a preset order to generate a fault propagation path state vector.

6. The substation multi-sensor data fusion monitoring and early warning method as described in claim 5, characterized in that, The process of performing state space attribution determination based on the fault propagation path state vector includes: Extract the unknown risk space boundary from the path state space benchmark library, substitute the coordinates of the fault propagation path state vector into the unknown risk space boundary to determine the spatial location, and generate a regional location identifier. In response to the location identifier of the area falling into an unknown risk area, the path status attribution determination result is marked as an unknown fault; In response to the fact that the area location identifier does not fall into the unknown risk area, the geometric center point coordinate set of the known fault categories in the path state space benchmark library is called to calculate the distance between the fault propagation path state vector and each cluster center point, and the known fault category corresponding to the minimum distance is selected as the judgment result.

7. The substation multi-sensor data fusion monitoring and early warning method as described in claim 6, characterized in that, The generated substation integrated monitoring and early warning status includes: In response to the path status attribution determination result being an unknown fault marker, an early warning status containing an unknown fault marker is generated; In response to the path status attribution determination result being a known fault category, a preset fault level mapping table is invoked, the corresponding risk level value is found according to the known fault category index, and the known fault category and the risk level are combined into a unified status description to generate a warning status containing the fault category and risk level.

8. A substation multi-sensor data fusion monitoring and early warning system, implementing the substation multi-sensor data fusion monitoring and early warning method as described in any one of claims 1 to 7, characterized in that, include: The state library construction module is used to define fault propagation paths based on the electrical topology of the substation, generate state vector sets based on historical sensor data, calibrate unknown risk space boundaries and known fault cluster centers based on the state vector sets, and construct a path state space benchmark library. The state vector generation module is used to collect real-time multidimensional sensor data for monitoring nodes on the fault propagation path, and generate a fault propagation path state vector based on the real-time multidimensional sensor data. The state space attribution determination module is used to call the path state space benchmark library, perform state space attribution determination based on the fault propagation path state vector, and obtain the path state attribution determination result. The monitoring and early warning module is used to generate the substation integrated monitoring and early warning status based on the path status attribution determination result.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and when the processor executes the computer-executable instructions, it implements the steps of the substation multi-sensor data fusion monitoring and early warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer-executable instructions are executed by the processor, they implement the steps of the substation multi-sensor data fusion monitoring and early warning method according to any one of claims 1 to 7.